The FORE-SCE model: Land Use Modeling ... - Farm Foundation

The FORE-SCE model: Land Use Modeling 

Parameterized with USGS Historical Data 

USGS EROS Data Center 

Terry Sohl 

Science Applications International Corporation (SAIC) 

Contractor to: 

USGS Center for Earth Observation and Science (EROS) 

Sioux Falls, SD 57198 

October 17th, 2007

USGS EROS Data Center


Land Cover Trends: 

Telling the story of 

Change

Applications of Land Cover and Land 

Cover Change Data 

NOAA temperature trends analysis 

Hale, R. C., K. P. Gallo, T. W. Owen, and T R. 

Loveland (2006), Land use/land cover change 

effects on temperature trends at U.S. Climate 

Normals stations. Geophysical Research Letters 

33: L11703 


Monthly Minimum Temperature Anomaly - Memphis, TN 


Monthly Minimum Temperature Anomaly ( o C) 

5.0 

4.0 

3.0 

2.0 

1.0 

0.0 

-1.0 

-2.0 

-3.0 

-4.0 

Dominant Change 

Period 

-5.0 

Jan-71 Jan-75 Jan-79 Jan-83 Jan-87 Jan-91 Jan-95 Jan-99 

Pre-change Pre change trend: -0.03 0.03 ± 0.06 °C/yr C/yr 

Post-change Post change trend: 0.15 ± 0.12 °C/yr C/yr


Cover Change Data 

USGS/NASA United States regional carbon dynamics 

Liu, J, S. Liu, and T.R. Loveland, 2006. Temporal evolution of carbon 

budgets of the Appalachian Forests in the U.S. from 1972 to 2000. Forest 

Ecology and Management (in press). 

Liu, J., S. Liu, T.R. Loveland, and T.L. Tieszen, 2006. Estimating carbon 

sequestration in agricultural ecosystems in the Western Corn Belt Plains of 

the United States. Global Change Biology (currently going through USGS 

review). 

Tan, Z., S. Liu, C. A. Johnston, T. R. Loveland, L. L. Tieszen, J. Liu, and 

R. Kurtz, 2005. Soil organic carbon dynamics as related to land use 

history in the northwestern Great Plains. Global Biogeochemical Cycles 

19: GB3011. 

Liu, S., T.R. Loveland, and R.M. Kurtz, 2004. Contemporary carbon 

dynamics on terrestrial ecosystems in the southeastern plains of the United 

States. Environmental Management 33: S442-S456. 


Total C Change (Mg C/ha/yr) 

2.5 

2 

1.5 

1 

0.5 

0 

-0.5 

-1 

Southeastern 

Plains 

1973 1978 1983 1988 1993 1998 

Year 

Ecoregion Carbon Dynamics 



2.5 

2 

1.5 

1 

0.5 

0 

-0.5 

-1 

Piedmont Northern 

2.5 

2 Piedmont 

1971 1976 1981 1986 1991 1996 

Year 


1.5 

1 

0.5 

0 

-0.5 

-1 

1973 1978 1983 1988 1993 1998 

Year


Cover Change Data – Conservation Issues 

USGS ARMI regional amphibian habitat analysis 

Price, S., M. Dorcas, A. Gallant, R. Klaver, and J. Wilson, 

2006. Three decades of urbanization: estimating the impact 

of land-cover change on stream salamander populations. 

Biological Conservation. 

USGS/FWS National Wildlife Refuge ecological integrity 

assessment 

Scott, J.M., Loveland, T.R., Gergley, K., Strittholt, J., and 

Staus, N., 2004. National Wildlife Refuge System: 

ecological context and integrity. Natural Resources 

Journal 44(4): 1041-1066. 


“The Influence of Historical and Projected 

Land Use and Land Cover Changes on Land 

Surface Hydrology and Regional Weather and 

Climate Variability” 

NASA ESE funded proposal 

T. Loveland, R. Pielke Sr., T. Sohl, L. Steyaert 

Conduct a series of regional experiments that collectively 

address the extent to which changes in specific regional 

land use and land cover characteristics, including the types, 

biophysical properties, and spatial configurations, affect 

surface hydrology and regional weather and climate 

variability. 

Sister USGS Prospectus funding 


Blue -- Primary 

Assessment Region 

Magenta -- Atmospheric 

Boundary 

Conditions 


Technical Components 

Land cover database development (pre-settlement, 

1920, 1992, and 2020) 

Biophysical parameter dataset development 

Sensitivity tests for summer months using 

RAMS/LEAF2/GEMTM 

Validation of model sensitivity 

Assessment of future regional climate vulnerability 


LULC databases -- LEAF2 (modified) 

classification scheme (Western Plains Application) 

• Oceans, Lakes 

• Ice Caps, Glaciers 

• Evergreen Needleleaf Forest 

• Deciduous Needleleaf Forest 

• Deciduous Broadleaf Forest 

• Evergreen Broadleaf Forest 

• Short Grass 

• Tall Grass 

• Desert 


• Evergreen Shrub 

• Deciduous Shrub 

• Mixed Woodland 

• Crops/Mixed Farming 

• Row Crop (Dryland) 

• Irrigated Crop 

• Bog or Marsh 

• Bare Ground 

• Urban and Built-up

Historical Land Cover Reconstruction 

A modified Kuchler’s 

Potential Natural Vegetation 

(PNV) is used to represent 

pre-settlement (~1650) 

~1920 chosen as it 

represents the period where 

agricultural activity reached 

a peak in much of the U.S., 

with eastern forests greatly 

reduced 

A cartographic modeling 

process used to reconstruct 

1920s land cover, using: 

PNV 

County Ag stats 

Census Data 

Other historical data sources 


Historical, 

Contemporary, and 

Future Land cover 

in the High Plains: 

Assessing the 

Consequences of 

Land Cover Change 

on Regional 

Weather and 

Climate Variability 


Pre-Settlement 1920 

1992 2020

Biophysical Parameter Construction 

Biophysical parameter datasets 

for coupled land-atmosphere 

modeling are based on mapped 

land cover classes. 

Aggregated time series of 

MODIS products (albedo, leaf 

area, veg. indices) used to 

improve the basis for 

prescribing these parameters 

for each land cover type. 

These data are combined with 

field and other data to construct 

final biophysical parameter 

data layers 

These serve as input to 

RAMS/GEMTM modeling 

system 


Projecting Land Cover Change: 

The FORE-SCE model 

FORE-SCE – FOREcasting SCEnarios of land use change 

Our goal is to: 

Provide scenario-based projections of change (3 scenarios per 

region) 

Project the proportions of each land cover in a region 

Site the cover on the lands with the highest potential for 

supporting that land use and cover 

Generate the clusters of cover and patterns of fragmentation so 

that the projected landscape represents a specific composition and 

configuration for each region. 


FORE-SCE: Model structure 

Issues of Scale -- Demand vs. Spatial Allocation module 

Demand Module 

Driven by macro-level socioeconomic drivers 

Drivers often not spatially explicit, not able to be mapped 

quantitatively 

Non-spatial...calculates area change for all land use types at the 

aggregate level 

Spatial Allocation Module 

Driven by lower-level socioeconomic and biophysical drivers 

Requires the establishment of quantitative relationships between 

LULC distribution and drivers 

Drivers MUST be mappable at the scale of analysis 

Output from Demand Module translated into spatially explicit 

allocation of land use change, based on a combination of 

empirical information, spatial analysis, and dynamic modeling. 


Land Cover Trends Data -- 

Our Ace-in-the-hole 

Use of Land Cover Trends project data 

We have empirical data on 

contemporary land cover change in the 

three study areas (change rates, patch 

characteristics, habitat fragmentation, 

drivers of change) 

Incorporation of Trends information -- 

Gives us an advantage that many 

modelers of LULC change don’t have. 

Existence of Trends data, 1992 NLCD, 

and 2000 NLCD offers the possibility 

for calibrating model results 


1992 

NLCD 

Climate 

(DAYMET) 

NED and 

Derivatives 

STATSGO 

Proximity to 

Transportation 

Population 

Data 

Socioeconomic 

Data 

Ag Census 

Data 

FIA 

Points 

Recode 


Modified 

NLCD 

Logistic 

Regression 

Interpolation 

(IDW) 

Stand Age 

(History) 

Elasticity 

Values 

Probability 

Surfaces 

(Land Cover(X)=1,19) 

Land Cover 

Trends Data 

Yearly Change 

Rates 

Prob(X)*Elas(X)* 

History 

Year(i)=1,58 

Layerstack- 

1992-2050 

Land Cover 

Patch 

Library 

Patch Size 

Characteristics 

TProb. X(i+1) 

History (i+1) 

Seed/Patch 

Placement 

Land 

Cover X(i+1) 

Final Land 

Cover (i+1)

FORE-SCE Model: Trends data and model 

parameterization 

Ecoregion-by-ecoregion model parameterization using Trends data 

Ecoregion specific parameters including change rates, change patch 

characteristics, “Elasticity” values, “Clumpiness” values 


Ancillary Data – Used in Logistic Regression 

Modified 1992 NLCD and all 

ancillary data compiled into 

one data stack 

50,000 points, minimum 

1,000 per land cover class 

Point land cover and 

corresponding ancillary data 

values imported into SAS 


Regression-based Probability Surfaces 


• Logistic regression 

used to analyze 

relationships between 

each land cover type 

and ancillary data sets 

• Probability surfaces 

constructed for each 

land cover type

Tracking Change through Time – 

HISTORY Variable 

Forestry activity in the Southeast results in an extremely dynamic landscape 

Over 20% of the Southeastern Plains ecoregion changed land cover type at least once 

between 1973 and 2000 

~75% of change in the Southeastern Plains is directly related to forest cutting 

Forest cutting cycles on loblolly pine plantations in the Southeast average 20-25 years 

Introduction of HISTORY variable that tracks “age” of every pixel in the SE study area 

Allows for much more realistic modeling of forest cutting cycle 


Establishing Stand Age – Forest Inventory 

and Analysis (FIA) Data 


Controlling Transition Probabilities 

“Elasticity” is an ecoregion-specific modifier to base probability values. It’s 

simply a number from 0 to 10 that gives the likelihood of a specific transition type 

occurring in that ecoregion. 

Elasticity for every possible land cover transition determined from Land Cover 

Trends data 

Powerful tool for control of scenario development. Depending upon the scenario, 

we can tightly or loosely restrict the types of land cover transitions that are allowed 

to occur. 


Calculating Total Probability (TPROB) 

Total Probability (TPROB) constructed prior to generation of change polygons 

TPROB = Baseline Probability * ELASTICITY * Function(HISTORY) * PAD 

ELASTICITY used both to modify probabilities based on likelihood of change, 

and to “zero out” probabilities for transitions we don’t want to occur 

Value 0-10, rescale to 0.0 to 1.0 for TPROB multiplier 

HISTORY used to zero out probability of new transitional (clear-cut) patches until 

forest stand age reaches minimum of 20 years 

PAD – Protected Areas Database 

Baseline Probability Total Probability 


Existing Cover 

Elasticity 

History 

PAD

Generating Change Polygons 

TRANSITION TYPE Mean Patch Size Standard Deviation 

Grass/shrub to Urban 7.10 5.84 

Grass/shrub to Mining 1.37 2.40 

Grass/shrub to Agriculture 57.14 114.22 

Grass/shrub to Wetland 2.88 1.08 

Agriculture to Urban 3.75 3.54 

Agriculture to Mining 6.84 8.41 

Agriculture to Grass/shrub 58.43 103.17 

Agriculture to Wetland 3.00 0.74 

Wetland to Agriculture 4.36 6.44 

Wetland to Wetland 3.76 8.85 


• Probability surfaces used in conjunction with 

Trends information to create change polygons 

• Use of “Seeds” planted on probability surfaces 

• Change patches assigned a realistic size range 

for that transition type 

• Uses lookup table of mean and standard 

deviation of patch size for each LC type, for 

each ecoregion (from Trends project) 

• Normal distribution of patch size modeled 

based on those parameters (simplification) 

• Random number (normal distribution) 

used to select patch size for each seed 

• DEMAND drives # of “seed” pixels 

• With normal distribution: 

# Seeds = TotalArea(lc x ) / Mean

Old Method 

New (improved?) Method 


Generating Change Polygons 

and FORE-SCE run time 

Once seed size is determined, a change patch 

centered on that seed is placed 

Old method – Region grow function into 

adjacent pixels of similar (or higher) 

probability 

Patches more “organic”, dependent upon 

probability surface characteristics 

Problem – Model Run times were 2-3 real days 

per yearly model iteration 

New method – Pre-established patch “library” 

of a limited number of patch configurations 

Assigned patch size for a seed determines pool 

of potential patch configurations 

Underlying probability surface still controls 

individual pixel placement 

“Canned” patches with less variability, but: 

1992 to 2050 model run (58 iterations) now 

takes just over 24 hours

Controlling “Clumpiness” of LULC change 


Probability surface guides 

placement of seed pixels for 

change 

Some land cover transitions 

are spread evenly across the 

landscape, some are more 

clumped 

Histogram breakpoints used 

to control “clumpiness” 

Ecoregion-by-ecoregion 

parameterization 

Highly dependent upon 

characteristics of individual 

probability surface 

Quantitatively determined 

from Trends data….???

“Footprint” of Change – 1992 to 2050 


RAMS/LEAF2/GEMTM Modeling 

RAMS/LEAF2 coupled modeling system 

is used to quantify the potential effects of 

LULC change on surface hydrology, 

weather, and climate variability 

Summer season (60-90 day duration) 

conducted, using each land cover data set, 

but with identical meteorological forcing 

data 

“Wet” year (1993, top right) and “Dry” 

year (2002, bottom right) averaged 

maximum monthly air temperature shown 

Top row – 1992 NLCD baseline run 

2nd – 4 th row – Differences from baseline 

S1 – Trends extrapolation 

S2 – Agricultural decline scenario 

S3 – Agricultural expansion scenario 



1992 to 2050 

Projected 

Change: 

Florida 


1992 to 2050 Projected 

LULC Change: 

Mobile, Alabama 



LULC Change: 

Montgomery, AL area 



LULC Change: 

New Orleans area 


1992 Conifer – Percent changed by 2050 


Red-cockaded Woodpecker – Conifer Stand Age 

Only bird in North America that regularly 

nests in cavities in living pine trees 

Longleaf Pine is preferred nesting tree 

Seek out Longleaf pine suffering from a 

fungal infection, “Red-heart disease” 

Causes inner wood to rot, become 

conducive for cavity building 

Longleaf pine and other pines don’t become 

susceptible to red-heart disease until an 

average age of 80-120 years old 

Also prefers contiguous patches of suitable 

habitat > 120 acres 

Even-aged, short rotation forestry 

management in the Southeast eliminates 

suitable nesting and foraging habitat 


Red-cockaded Woodpecker – Conifer Stand Age 

RCW Family Clusters, 1990-1994 

15 Groups of 100+ Family 

Clusters, all but 1 on public land 


Projected 2050 Conifer Stand Age and Historical RCW Range 

Blue: Stand Age 75 

Crosshatch: Historical Range

Great Plains -- 1992 to 2020 

Change (Agriculture 

Expansion Scenario) 


Scenario Comparison -- SW Kansas 


South Dakota -- 1992 to projected 2020 Change 

Northern Great Plains – Study Area 


FY2008 development: Northern Great Plains – 

Biofuel-driven Agricultural Change 


Primary goal – To evaluate the 

effects of an expanded agricultural 

base for biofuels and concurrent 

changes in climate on ecosystem 

sustainability across the Northern 

Great Plains 

Development of credible landchange 

scenarios to project 

alternative landscape futures 

through 2050 

Analyze results to estimate effects 

on ecosystem processes and 

services.

Northern Great Plains Biomass for Biofuel: 

Key Research Questions 

What landscape patterns are likely to result from an expanded 

biomass-for-biofuel economy and what are our estimates of the 

uncertainties? 

What are the environmental consequences? 

What are the full economic costs and benefit of biomass 

production for energy, including agricultural sector profitability? 

How will projected climate change impact agricultural production 

and profitability? 

How will projected climate change impact the provision of 

ecosystem services? 

What are the feedbacks among land-use change, economic and 

policy drivers, climate, biophysical processes, and a variety of 

ecosystem services? 


FY2008 development: Scenarios 

Corn for ethanol – Large-scale production already underway in 

the Northern Great Plains, with continued expansion 

Soybeans for biodiesel 

Grasses for cellulosic ethanol 

Switchgrass 

Miscanthus 

Mixed Prairie Species 

Climate – IPCC-defined scenarios 

No change, Low-IPCC, High-IPCC 


FY2008 development: DEMAND 

DEMAND in current FORE-SCE 

applications has largely been 

based on extrapolations of 

historical (L.C. Trends) data 

Priority: Building a DEMAND 

module that instead is driven by 

policy, economic costs and 

returns, and biophysical 

conditions 

Prerequisite – Quantifying 

relationships between these 

drivers and actual land cover 

change 


Public policies 

Biofuel 

demands 

Commodity 

prices 

Climate 

Change 

Drivers 

and 

Feedbacks 

Drainage 

subsidies 

Site 

Conditions 

Irrigation 

incentives 

Land Cover Hectares 

Corn 18,500,000 

Soybeans 12,200,000 

Switchgrass 6,194,000 

Hay/Pasture 23,529,000 

Etc. Etc.

FY2008 development: Parcel (field) based 

modeling 

CLU 

NASS CDL 


MODIS NDVI 

USDA – Common Land Unit 

(CLU) 

Smallest unit of land that has a 

permanent, contiguous 

boundary, a common land cover 

and land management, a 

common owner and a common 

producer in ag land associated 

with USDA farm programs. 

MODIS 16-day Vegetation 

Index composites – Used to map 

crop type, augment 2001 NLCD 

classification 

NASS Cropland Data Layer 

(CDL) used to validate MODIS 

crop-type information

FY2008 Development – Model integration 


FY2010 and Beyond 

USGS plans for national land-cover monitoring 

NLCD updates every 5 years 

NLCD change products 

Integrate Trends expertise 

And…land cover projection component?

The FORE-SCE model: Land Use Modeling ... - Farm Foundation

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?